01656602

Component

This paper proposes an econometric framework to model the multivariate panel crash by type count data. The point of emphasis is that modeling multivariate panel count data has superior econometric benefits in producing more efficient parameter estimates compared to the ones arising from the cross-sectional concept. Within this context, the authors considered the intertemporal correlations of a given crash types among different years of observations, while simultaneously accounting for inter-crash type. The authors developed two flexible models to address this joint problem: A Multivariate Panel Poisson Gamma Copula (MVPPGC) and Multivariate Panel Copula-Copula (MVPCC) model. The applications of the proposed models are demonstrated using a balanced panel crash dataset for three years (2005-2007). A total of 274 multilane freeway segments in the State of Washington, USA were analyzed. The empirical results suggest that the Frank copula statistically outperformed other copula types for fitting the intertemporal correlation among the years of observations of each crash type. MVPCC model offers a better prediction of the crash-type count than MVPPGC model, since it more accurately represents the variance-covariance structure.

This paper was sponsored by TRB committee ABJ80 Standing Committee on Statistical Methods.

18-01170

English

7p

2018

Transportation Research Board 97th Annual Meeting

Digital/other

References (21) ; Tables

Crash characteristics; Crash data; Econometrics; Mathematical models

Copula-based models

Washington (State)

Economics; Highways; Safety and Human Factors

Transportation Research Board Annual Meeting 2018 Paper #18-01170

TRIS, TRB, ATRI

Jan 8 2018 10:17AM